Source: Cao, Jiang, Yang & Zhang (2023) — “How to Talk When a Machine Is Listening: Corporate Disclosure in the Age of AI”
The part that bothers me is how calm the language sounds. It makes the risk feel distant, when the real problem is already sitting inside today’s systems and decisions.
There’s a feedback loop in corporate disclosure that most people don’t think about. But increasingly, the “investors” doing the reading aren’t human — they’re algorithms and AI systems operated by quantitative funds and institutional investors. And the question this paper asks is whether firms know this, and whether they respond. The researchers find that as AI-equipped investors acquired larger stakes in firms, and as algorithmic download activity of SEC filings increased, firms began producing filings with more machine-friendly linguistic properties. Specifically, negative sentiment in 10-K filings declined over time in ways that track the rise of machine readership — and the effect is stronger for firms with more to gain from sentiment management external financing needs and less to lose lower litigation risk. The researchers use two events as quasi-experiments: the publication of the Loughran and McDonald 2011 sentiment dictionary, which became a widely used tool for automated filing analysis, and the release of BERT in 2018, a neural language model that became a foundation for AI-based text analysis in finance. In both cases, firms appear to adjust their filing language in ways consistent with gaming the algorithmic signals those tools produce.
In plain English, that is why the result matters beyond the chart. It changes where people should look, what they should question, and which comfortable assumption probably needs to be retired.
For leaders, the lesson is simple: if the risk timeline changes, the attention timeline has to change too. Waiting until everyone agrees it is urgent is usually how organisations arrive late.